Windows 11 is quietly gaining a feature that could hand you direct control over one of the most important hardware decisions on next-generation PCs: how much shared memory gets set aside for graphics and AI work. It sounds like a niche setting, but for the wave of unified-memory machines arriving this fall, it could be a meaningful lever.
References to this capability have turned up in recent Windows 11 preview builds, pointing to a hidden feature Microsoft is testing under the internal name IntelligentCarveout. It would let Windows reserve a fixed chunk of unified memory for your GPU and AI accelerators — and, crucially, pull that memory out of reach for your other applications.
What’s hiding in the build?
Windows Latest first spotted the references in internal NVIDIA documentation, and Windows Insider watcher phantomofearth confirmed that the same strings show up in actual Windows 11 builds. The build in question is 29648.1000, released August 17, which includes the hidden IntelligentCarveout feature carrying feature ID 61121285, alongside a new file called SettingsHandlers_UnifiedMemory.dll.
The build also contains several telling strings. Microsoft’s code references “Reserved memory for accelerators” and “Memory for graphics and AI acceleration.” One string spells out the intent plainly: “Let Windows reserve additional unified memory for graphics and AI intensive games and applications. Reserved memory is not available for other applications.”
That last sentence is the key detail. Unlike the “Shared GPU Memory” you can already see in Task Manager — which is really just a ceiling Windows can dip into when the workload calls for it — this looks like an actual carve-out. The reserved memory would be locked away from other programs entirely.
Based on the evidence in the build, the final settings page would likely live under Settings > System > Advanced, though Microsoft hasn’t shown what the actual control will look like. It could be a slider, fixed presets, or workload-based profiles. As of now, the build’s release notes say nothing about the feature, which is typical for something still in the experimental stage.
Why unified memory changes the game
Unified memory is a single, high-bandwidth pool shared between the CPU, GPU, and other accelerators, rather than the traditional split between system RAM and dedicated video memory. Apple popularized the approach with its M-series chips, and it has become especially valuable for running local AI models, which need to shove huge amounts of data through that shared pool in one place.
On a typical laptop with a discrete GPU, memory stays in two buckets: RAM for the CPU and VRAM for the graphics card. But platforms like NVIDIA’s RTX Spark, AMD’s Ryzen AI Max “Halo” series, and other upcoming designs put everything on one chip, sharing memory across the board. That’s the direction the industry seems to be heading.
RTX Spark is the poster child here. It packs a 20-core Grace Arm CPU and a Blackwell GPU with up to 6,144 CUDA cores, plus AI acceleration and up to 128GB of unified memory, all on a single chip. That’s a lot of shared RAM to manage intelligently, and it’s exactly the kind of hardware a memory carve-out setting was built for.
Here’s where the Windows approach diverges from Apple. macOS doesn’t give MacBook owners any direct way to adjust how much unified memory is reserved for graphics or AI. On Windows, Microsoft appears to want to hand that control to you, which makes sense given that Windows is also the leading platform for gaming, not just productivity. A gamer who wants to run a local model in the background shouldn’t have to pick between the two.
Running a big local model changes the math
Think about what actually changes day to day. When you’re running a large local AI model, that model lives in unified memory. Reserving more of it for AI and graphics means the model has room to breathe, but it also means less is available for your browser, your editor, or whatever else you have open at the time.
When you’re just gaming or doing regular work, you’d want to leave as much of that memory free as possible. A manual control lets you tune that trade-off instead of letting the OS guess, and it could be the difference between a smooth session and a system that chokes because the model and your apps are fighting over the same pool.
Microsoft is already laying the groundwork for this kind of hardware. The company is testing what it calls Workload Profile Scheduling, a rebuild of Windows 11’s task scheduler, memory management, and power management specifically tuned for RTX Spark. Regular Windows scheduling simply wasn’t built for a chip like that running on a portable laptop.
NVIDIA has also already shipped a developer-preview GeForce driver, version 616.00, for Windows 11 on Arm. That driver confirms two N1X configurations, one with 6,144 CUDA cores and one with 5,120. Leaked Geekbench 7 scores for the RTX Spark N1X chip have already reportedly beaten AMD’s Ryzen AI Max+ 395 and Intel’s Core Ultra X9 388H in multi-core performance, on hardware that isn’t even final yet.
What this means for you
For most Windows users today, this feature is a long way from your screen. Unified-memory PCs are still rare, and the ones shipping this fall — RTX Spark laptops and mini PCs from Microsoft, Lenovo, ASUS, Dell, HP, and MSI, with Acer and GIGABYTE to follow — are aimed at power users and developers, not casual gamers. The Surface Laptop Ultra, Microsoft’s RTX Spark answer to the MacBook Pro, and Lenovo’s Yoga Pro 9n are among the first machines expected to use the platform.
But the setting doesn’t appear to be exclusive to RTX Spark. It’s found in the Experimental “Future Platforms” branch, which suggests other unified-memory chips could benefit once they land. If you ever buy a high-end unified-memory laptop, this could be the kind of control that helps you get the most out of it, whether you’re training a small model locally or pushing the highest settings in a demanding game.
The honest takeaway: this is early-stage engineering, not a feature you can use this week. Microsoft hasn’t commented on IntelligentCarveout, and the build’s release notes say nothing about it. Treat it as a signal of where Microsoft is heading, not a promise. If you’re running the affected build, expect the setting to remain hidden until Microsoft decides it’s ready for a wider audience.
How to try it now
If you’re on the Experimental preview channel and want to poke around, the feature can potentially be enabled with ViVeTool using feature ID 61121285. That’s a third-party toggle tool aimed at Insiders, and enabling hidden features can make your system unstable — so proceed at your own risk and only on a machine you don’t rely on for work.
For everyone else, there’s nothing to do yet. Keep an eye on the Windows Insider Blog and Microsoft Learn’s experimental future-platforms release notes, where Microsoft will hopefully announce this once it’s ready. The verified build details live in the experimental future-platforms release notes for build 29648.1000, so that’s the authoritative place to check for updates.
Bottom line: Windows 11 unified memory control is taking shape in the background, aimed at a hardware wave that’s about to arrive. It’s a small setting on paper, but for anyone running local AI or pushing graphics on a unified-memory machine, it could be exactly the kind of control that makes the difference between a system that works for you and one that works against you.
Source: Windows Latest
Over to you: If Windows 11 lets you carve out unified memory for AI, would you lean toward reserving more for local models, or keep it all free for gaming?
